home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math.stat
- Path: sparky!uunet!utcsri!torn!watserv2.uwaterloo.ca!mach1!hunt7551
- From: hunt7551@mach1.wlu.ca (len hunt 9208 U)
- Subject: HEYWOOD CASES
- Message-ID: <BxB5u9.KpG@mach1.wlu.ca>
- Organization: Wilfrid Laurier University
- Distribution: na
- Date: Fri, 6 Nov 1992 18:54:08 GMT
- Lines: 28
-
- I was wondering if someone could help me with a problem I have encountered.
- While completing a number of principal axis factor analyses several have had
- problems with heywood cases, i.e., the final estimate for a communality
- exceeds 1.00.
-
- The problem involves the handling of this situation. First, I do not wish
- to alter the data in any way or try different variables; there is a good
- reason for this but for sake of brevity I shall not disclose it. I have
- read and heard that one can set the communality of a heywood case to 0.99
- instead of the >1.00 that is found. However, correct me if I am wrong, the
- squared factor loadings for this variable by all factors will still exceed
- one. Therefore, I have stopped the iteration at the point before this
- heywood case arises instead of setting the value equal to 0.99. It is true
- that the final convergence criteria for the communalities is not met by
- doing this, but I think it is a better way to handle these cases.
-
- I would appreciate hearing your comments upon stopping the iteration
- beforehand. Please send your message directly to me, as there is already
- enough to read on these news lists.
-
- Thanks Sincerely,
-
- Len Hunt
-
- Wilfrid Laurier University
- Waterloo, Canada
-
- E-mail ...hunt7551@mach1.wlu.ca
-